Polling Labs · Free tool
Margin of Error Lab
That “±3 points” footnote decides whether a headline is news or noise. Move the numbers and see how much it really changes.
1 · The basics
Your poll's margin of error
±3.1
percentage points
The true value is plausibly between 44.9% and 51.1% — a window 6.2 points wide.
2 · Why it matters
Is this race actually a lead?
Uses the sample size and confidence level from above.
Gap: 3 points. Margin on that gap: ±6.0 — about 1.9× the headline ±3.1.
The gap is real-looking but within its own margin.
3 · Diminishing returns
More people, slower gains
Quadrupling the sample only halves the margin. Going from 1,000 to 4,000 takes you from ±3.1 to ±1.5.
4 · Subgroup shock
Zoom in on a group
±8.0
vs ±3.1 for the full poll — 2.6× wider. Readable, still check the margin.
5 · Same poll, 20 draws
Run the same poll 20 times
If the true value were 48%, here is what 20 separate polls of 1,000 people might report. The shaded band is ±3.1.
1 of 20 landed outside the band. At 95% confidence, expect about 1.
Illustrative simulation of random sampling only.
6 · The fine print
What the margin of error doesn't cover
- • Leading or confusing question wording
- • People the poll never reached
- • Who chose not to answer
- • Weighting choices and their extra variance
- • Fake, duplicate, or bot respondents
- • Opt-in samples that aren't truly random
The stated margin assumes a clean random sample. Real error is usually larger. PolitiMind works on the parts it ignores — verifying each participant is a unique real person and recording how results were produced.
Questions
What is a margin of error in a poll?
It is the range around a poll result where the true value plausibly sits, from random sampling alone. A 50% result with ±3 points at 95% confidence means roughly 47% to 53%.
Why does a 3-point lead with a ±3 margin not mean a clear lead?
The uncertainty on the gap between two candidates is larger than the headline margin — close to double when both shares come from the same poll. A 3-point gap can easily be noise.
Does a bigger sample always fix it?
Larger samples shrink the margin, but slowly: quadrupling the sample only halves it. And the margin does not cover bad wording, unreachable groups, nonresponse, or fake respondents.
Why are subgroup results less reliable?
A subgroup is a smaller sample. If 15% of a 1,000-person poll is 18–29, that group's margin is about ±8 points, not ±3.